Google's Closely Guarded Secret -- The Tensor Processing Unit

Google is making technological leaps. From simple mobile up to its processing units -- Google has made a lot of difference -- even significant difference. The company has had a successful stride toward planning, making, testing and eventually manufacturing autonomous cars. This time, it will be another face that it is trying to make beyond a person's perception.

According to a post via PC World, Google has just unveiled its Tensor Processing Unit. It is a custom chip that advances computing performance by three generations. Google CEO -- Sundar Pichai -- said in the company's I/O developer conference that Tensor Processing Unit delivers higher performance per Watt than all commercially available Graphic Processing Units.

In addition, Google mentioned that the company has been running TPUs inside its data centers for more than a year now. The TPU delivered an order of magnitude better. They added that this is so futuristic that it would even equalize a three generational futuristic processor.

The TPU is tailored to machine learning applications. It requires fewer transistors per operation. Because of this TPU, they will be able to use more operations per second. Additionally, they will be able to use more sophisticated and powerful machine learning models. This will eventually lead to more intelligent result for the users.

According to a post via Computer World, the TPU is in production use across Google's cloud. The company has thousands of these chips in use. These are fit to Google's data center racks.

In addition, Google does not want to sell these chips especially to its greatest competitors. It does not want to give its rivals access to the chips.

Patrick Moorhead, an analyst, expects that these chips will be used for inferencing. It is a part of machine learning operations that does not require much flexibility. It is also reported that the company will reveal more about TPUs in the coming months.

© 2021 iTech Post All rights reserved. Do not reproduce without permission.

More from iTechPost